Marketing measurement that survives scrutiny: ROI, mix modelling and useful dashboards
Marketing measurement fails when it optimises for reporting rather than deciding. Build a hierarchy — business outcome, intermediate effect, activity — then use experiments for causal claims, mix modelling for allocation, and dashboards for exceptions. Fewer numbers, defined once, reviewed on a rhythm, beats a wall of charts nobody acts on.
What is marketing mix modelling?
Marketing mix modelling is a statistical approach that estimates the contribution of each marketing channel and external factor to a business outcome over time, using aggregate historical data. It supports budget allocation decisions across channels and is privacy-durable because it does not rely on individual-level tracking.
Build the hierarchy before the dashboard
Most measurement debates are structural: a channel metric is being compared against a business outcome, or an intermediate effect is being treated as proof of revenue. A three-layer hierarchy resolves most of it.
The top layer is business outcomes — revenue, margin, customer acquisition and retention, price realisation. These are what the organisation is actually trying to move, and marketing contributes to them jointly with product, pricing and sales.
The middle layer is intermediate effects with an evidenced link to those outcomes — qualified pipeline, share of search, awareness and association, repeat rate, average order value. These are marketing's genuine accountability zone.
The bottom layer is activity — impressions, clicks, cost per click, open rates, engagement. Useful for optimisation inside a channel, and actively misleading in a board pack.
Use the right instrument for the question
Different questions need different tools, and mismatching them is the most common analytical error we encounter.
Attribution's honest role
Multi-touch attribution has been oversold and is now sometimes over-dismissed. Its legitimate use is directional optimisation within digital channels where tracking is intact. Its illegitimate use is settling questions of cross-channel budget allocation or proving brand contribution, because it can only see what it can tag and systematically over-credits the last, cheapest touch.
Practically: keep attribution for in-channel decisions, use experiments and modelling for allocation, and stop quoting attributed ROAS as though it were incremental return. The credibility of the whole measurement function depends on that distinction being visible to finance.
Dashboards designed for decisions
The test for any dashboard is whether a specific person, at a specific cadence, changes a specific decision because of it. Everything that fails that test is decoration with a maintenance cost.
That leads to a small set: an executive view of outcomes and modelled contribution reviewed monthly; a marketing operating view of intermediate effects and pipeline reviewed weekly; channel views owned by their teams; and an exception layer that alerts rather than waits to be visited.
Two design rules earn their keep. Every metric carries its definition and source on hover. And every chart shows a comparison — target, prior period, or forecast — because a number without a reference point cannot be judged.
Reporting brand and performance in one pack
Boards lose patience with marketing when brand and performance are blended into a single claim. Report them side by side with explicit timescales and explicit uncertainty: response outcomes for the period with incrementality caveats stated, brand indicators as trend lines over four to eight quarters, and modelled contribution with confidence ranges.
Stating uncertainty increases credibility rather than reducing it. Finance teams are used to ranges; they are suspicious of unqualified precision, and rightly so.
Reporting-led measurement vs. decision-led measurement
Frequently asked questions
How much data is needed for marketing mix modelling?
Typically two to three years of weekly aggregate data across channels, with meaningful variation in spend. Insufficient variation is the usual blocker: if budgets have been flat and identical by channel, the model cannot separate their effects.
What is the difference between attribution and incrementality?
Attribution allocates credit among observed touchpoints for conversions that happened. Incrementality estimates what would not have happened without the activity, by comparing exposed and unexposed groups. Only the second answers whether the spend was worth making.
How do we measure marketing when tracking is restricted?
Lean on aggregate and experimental methods: geo holdouts, mix modelling, share of search, direct and branded demand trends, and first-party outcome data joined server-side. These were always the more robust methods; privacy change simply removed the easier alternative.
How many metrics should a marketing dashboard show?
For an executive view, five to eight, each with a comparison and a definition. Anything more is browsed rather than read, and browsing does not produce decisions.
A short diagnostic conversation is usually enough to tell you whether there is a real opportunity here — and what it would take.